Dense frequency maps by Structure Tensor and logarithmic scale space: application to forensic fingerprints

Increasingly, reliable absolute frequency and orientation maps are needed, e.g. for image enhancement. Less studied is however the mutual dependence of both maps, and how to estimate them when none is known initially. We introduce a logarithmic scale space generated by the trace of Structure Tensor to study the relationship. The scale space is non-linear and absolute frequency estimation is reduced to an orientation estimation in it. We show that this offers significant advantages, including construction of efficient estimation methods, using Structure Tensor yielding dense maps of absolute frequency as well as orientation. In fingerprints, both maps can successively improve each other, combined in an image enhancement scheme via Gabor filtering. We verify that the suggested method compares favorably with state of the art, using forensic fingerprints recognition as test bed, and using test images where the ground truth is known. Furthermore, we suggest a novel continuous ridge counting method, relying only on dense absolute frequency and orientation maps, without ridge detection, thinning, etc. We present new evidence that the neighborhoods of the absolute frequency map are useful attributes of minutiae. In experiments, we use public data sets to support the conclusions.